Every AI product-image tool opens with the same gallery: a perfume bottle on wet stone, a trainer floating over a soft gradient, a ceramic mug in morning light. The samples look flawless because someone chose them. None of them is your product.
What the gallery leaves out shows up after you've paid: an export smaller than your shop asks for, a grey halo around a glass edge, a label redrawn in the wrong red, and a licence page you never opened. By then you have a subscription and a folder of scenes you can't use.
The takeaway: shortlist two tools, then run six checks during the trial, on your own product photos, before you subscribe. The checks cover product fidelity, edges and shadows, export, control, licence terms and batch consistency. A tool that fails one of them in the trial is cheaper to drop now than after the first invoice.
For the shortlist itself, one starting point is Eric Salo Reviews. It publishes short, first-person reviews of online tools, software, courses and services, and its AI coverage includes AI product-image tools. Read what a review says the product is and who it suits, add the tools your own search turns up, and narrow the field to two; the checks below decide between them.
Sponsored by Eric Salo Reviews. The checklist works for any AI product-image tool; no tool is named here.
Prepare the trial before you open either tool
If the trial is time-limited, do the setup before you sign up, not after. Gather four kinds of product photo from your own catalogue, each one a known trouble case:
- Something reflective or transparent: glass, chrome, a glossy bottle.
- Something with fine edges: fur, fringe, mesh, a plant, loose thread.
- Something with a colour you know exactly: a brand red, a dyed fabric, a printed label.
- Something whose size matters: an item buyers misjudge from photos, like a small jar or a large tote.
Keep the original, unedited photo of each one as your reference. Then open your shop's own image-spec page and the spec page of each ad platform you use, and write down what they ask for: dimensions, aspect ratio, format, file size and background rules. Specs change, so take them from the platform's own page, not from memory.
Copy the trial sheet below into a spreadsheet, one column per tool.
Six checks for an AI product image tool during the trial
In the US, the National Institute of Standards and Technology publishes the AI Risk Management Framework, a voluntary framework released in January 2023. It treats trustworthiness as something to weigh at every stage of an AI product's life, from how it is designed and built to how people use it and evaluate it. A buyer's trial sheet is a small piece of that last stage: evaluation, done on your own material.
Run every check on both shortlisted tools with the same photos, and judge the downloaded file, not the preview in the browser.
1. Does the output still show your real product?
Put each generated scene next to the reference photo of the item you ship and compare three things.
- Colour. Is the brand red still your red, or did it drift warmer or duller? Check the printed label and any logo on the product too.
- Shape. Did the tool smooth a seam, change a cap, straighten a handle, add a button or redraw the lettering?
- Scale. Does a small jar sit in the scene like a small jar, or have the props and setting made it read like a vase?
The point of a product image is that the buyer recognises the item when the parcel arrives. Pass: all three match, or you can correct the difference in your own editor without repainting the product. Fail: the tool changes the product in ways you would have to rebuild by hand on every image.
2. Edges and shadows at full size, not in the thumbnail
Whether the tool is a full scene generator or an AI background generator for products, cutouts fail at the edges, and thumbnails hide it. Open the export at 100% zoom and trace the outline of the product. Look for a halo of the old background, a soft fringe on hair or fabric, a bite taken out of the product, or glass that has turned opaque.
Then look at the shadow and the light together. The contact shadow should sit where the product touches the surface, and the cast shadow should fall away from the scene's light. A product with no contact shadow, or a highlight that disagrees with its shadow, reads as pasted in, even when the viewer can't say why.
Start this check with the reflective and fine-edged photos. They fail first.
3. Export sizes and formats
Set the spec list you wrote against what each AI product photo tool actually exports:
- Does the trial export at the same size and quality as the plan you would pay for, or is it smaller or watermarked?
- Can you export each size on your list without upscaling, and in the formats you need, including a transparent background where your shop wants one?
- Is there a monthly export allowance, and does a re-run or a resized copy count against it?
- Does the file carry the colour profile your shop expects?
If the trial exports smaller than your shop needs, it can't show you a paid export. Ask the vendor for a full-size sample of your own product, or read its refund terms before you buy.
Once the file is out of the tool, the usual export rules still apply. This blog's guide to exporting images for the web without losing quality covers format, dimensions and compression for the final file.
4. Control: masks, re-runs and one change at a time
A tool you can't steer hands you a new gamble on every click. Check four things:
- Masks. Can you protect the product so only the background and lighting change?
- Re-runs. Can you regenerate the same scene and get a close variation, instead of a different picture each time?
- One variable at a time. Can you change only the surface, only the light or only the angle, and keep everything else? The method is the same as changing one variable at a time in an AI design prompt: if three things change at once, you can't tell which instruction did it.
- Hand-off to your editor. Can you take the result into your own editor, ideally with product and background on separate layers, so the last round of product image editing is yours?
Pass: you can fix a bad shadow or a colour drift without starting over. Fail: every correction means regenerating the whole image and hoping.
5. Commercial-use and licence terms
Before you use a generated scene for a client or in an ad, read the tool's own licence or terms page for generated images. Look for answers to these questions:
- Is commercial use allowed on the plan you would buy, and is the trial any different?
- Can you use the images in client work and in paid ads?
- What do the terms say about the product photos you upload?
- What happens to images you have already made if you cancel?
If the page doesn't answer a question, ask the vendor in writing and keep the reply before any scene goes to a client.
6. Batch consistency across a catalogue
A single strong image is a sample. A shop needs the same look across many products. Run a small batch: one scene setup, several different products from your trouble list.
Compare the results side by side: light direction, camera height, shadow softness, background colour, crop and the margin around the product. Then check whether you can save the setup as a preset or template, and whether running the same product again the next day gives you the same look. A tool that produces one good image per session leaves the matching to you, by hand, on every product after it.
A trial sheet you can copy
Fill in one column per shortlisted tool. Mark each check pass or fail, and write one line on why.
| Check | What to do in the trial | Pass if | Fail if |
|---|---|---|---|
| 1. Real product | Compare each scene with the reference photo | Colour, shape and scale match, or you can correct them in your editor | The product is redrawn in ways you would fix by hand every time |
| 2. Edges and shadows | Inspect the export at 100% zoom | Clean outlines; contact shadow and light agree | Halos, fringes, opaque glass, a floating product |
| 3. Export | Compare exports with your shop's and ad platforms' spec pages | Every size and format on your list, at paid-plan quality | Trial exports differ from the paid plan, or a size you need is missing |
| 4. Control | Mask, re-run, change one variable | You can fix one thing without regenerating everything | Every change gives you a different image |
| 5. Licence terms | Read the tool's own terms page | Commercial and client use clearly covered on your plan | Terms are missing, unclear or narrower than your use |
| 6. Batch | Same scene, several products | Light, crop and background match across the set | Each product comes out with a different look |
Two tools, six rows. If only one passes every row, subscribe to it. If both do, pick the one whose exports need less correction in your own editor. If neither does, drop any tool that fails check 1 or check 5, because an editor can't fix those, and keep the one whose fails you can correct in your editor. If both are dropped, go back to the shortlist before you pay.
Building the shortlist with Eric Salo Reviews
A review on Eric Salo Reviews says what a product is, what stands out and who it suits, with a link to the product. Eric Salo personally explores and analyses the tools he reviews.
For this job, read what each review says about who the product suits, and hold that against your own work: a solo seller with a small catalogue, a freelancer preparing images for clients, or a shop that needs one scene across many products. Drop any tool whose fit is clearly someone else's, set the rest against the tools your own search turns up, settle on two, and start both trials with the same photos and the sheet above. Its AI product-image and other AI tool reviews page sets out the kinds of AI tools it covers, from prompt builders to product-image tools. If a tool you are weighing isn't covered, readers can suggest tools for future coverage through its contact form.
The trade-off: Eric Salo Reviews places affiliate links on some of its pages, and it may earn a commission at no extra cost to you. An AI image tool review is not a spec sheet either, so confirm export limits and licence terms on the vendor's own pages, and let your trial sheet make the final call.
FAQ
Can I use an AI-generated product scene in ads?
Check the tool's commercial-use licence terms and the ad platform's own rules before you run it, and keep the product's colour, shape and scale true to the item. Keep the reference photo with the generated file, so whoever approves the ad can compare the two.
Why does the product colour shift in an AI scene?
Unless the product is masked, the tool may redraw it along with the scene, so the colours are its interpretation, not your photo. Compare every scene against a reference photo, correct the colour in your own editor, and check the colour profile on export. If the tool lets you mask the product, use the mask and compare again.
What do the affiliate links on Eric Salo Reviews mean for my choice?
They don't change your price: Eric Salo Reviews says some of its links are affiliate links, and it may earn a commission at no extra cost to you. They do mean a review is the start of your shortlist, not the decision. Confirm plan and licence details on the vendor's pages, and let your own trial judge the output.
Next step
Pick the product photo that gives you the most trouble, the glass bottle or the fabric with the exact dye, and make it the first image in both trials. Then open EricSaloReviews.click, read what its AI tool reviews say about who each tool suits, add what your own search turns up, settle on two candidates, and run the six checks before the first invoice.